Metro 2033 Performance review

[mfenn]

okay it makes more sense now.

Hey, I'm not completely off my rocker yet!

 

[legcramp]

Toyota said Depth of View and Tesselation need to be turned off on PC? Doesn't Xbox run Tessellation standard?

haha.....

 

[MrK6]

I am curious to see some cpu usage as I am wondering if this game is actually using a quad. also I saw someone mention in another forum that the game uses lots of vram and that 512mb doesnt cut it and even 1024mb can be exceeded at 1920 with the right settings.

I'm running 2560x1600 with everything maxed except for AA and Advanced DoF and I have yet to hit a VRAM wall. The game runs very smoothly and doesn't "skip" while it loads environments like STALKER (very similar engine). Either way, the GPUs seem to be maxed before the VRAM is, unless the manifestations of VRAM limitations are somehow different in this game.

 

[1h4x4s3x]

A $300 video card made in 2009 is getting less FPS than a $200 console built back in 2005?

This is the kind of stuff that makes me wonder about the future of PC gaming? Maybe we will start to see ATI entering the HPC market?

Is that some failed trolling? Because it doesn't make any sense all.

 

[evolucion8]

I just played the first two levels of Metro 2033 in my PC and the CPU usage is very erratic, it never exceeds 40% in any of the cores, it uses more the core 1 and 3, and the other ones runs with lower usage, plus it has so many spikes with ups and downs that I feel that the game isn't CPU hungry. I have PhysX disabled, in the 2nd level where the guys are drinking, with Crossfire disabled it runs at 26fps average, with it enabled, it runs at 40fps enabled, seems that's the game isn't very optimized yet, even though it can put a nice load on both GPU's, the game stutters a bit, but far from being unplayable, it is maxed plus 1280x1024, 4xFSAA, DX10 and everything on highest, odd that ATi Tray Tool states that the game is running at DX11, probably not all the DX11 effects are enabled since DX11 is a superset of DX10.

 

[Dorkenstein]

What kind of hardware is required to comfortably use DX 11 mode with tesselation? Or is it beyond just about all setups? Thanks.

 

[toyota]

What kind of hardware is required to comfortably use DX 11 mode with tesselation? Or is it beyond just about all setups? Thanks.

5850 crossfire or better.

 

[Dorkenstein]

So it requires more than one gpu? Or at least one that's better than a 5850?

 

[Keysplayr]

So it requires more than one gpu? Or at least one that's better than a 5850?

19x12 AAA 16xAF Tesselation on, DoF on, DX11 All settings high is not very playable on this 5870. I'm getting consistent low 20's. So I think he was right on with Dual 5850's for these type of eye candy settings. I can play with the settings and lower something here and there to get it to a playable level. Still trying different things.

 

[Dorkenstein]

What would have to be sacrificed to run it with MSAA 4x? I switched from DX11 with DOF and Tess OFF, High settings, AAA to MSAA4x, DX10, High Settings. Haven't tried getting into a big fight with the latter settings though.

 

[MrK6]

What kind of hardware is required to comfortably use DX 11 mode with tesselation? Or is it beyond just about all setups? Thanks.

Depends on the resolution. For 1920x1200 and under you're fine with a 5870 or overclocked 5850. One of my 5850's @ 900/1250 plays 1920x1200, DX11, Tesselation on, everything else maxed at ~40FPS, it might dip down to 25FPS in the stations. Even then, one really doesn't notice it because the game plays so smoothly. If I go up to 2560x1600, especially with my increased field of view, I have to enable Crossfire.

19x12 AAA 16xAF Tesselation on, DoF on, DX11 All settings high is not very playable on this 5870. I'm getting consistent low 20's. So I think he was right on with Dual 5850's for these type of eye candy settings. I can play with the settings and lower something here and there to get it to a playable level. Still trying different things.

That's more due to DoF than Tesselation. Tesselation I've noticed gives about a 10-15% FPS hit, where as DoF cuts it in half almost.

What would have to be sacrificed to run it with MSAA 4x? I switched from DX11 with DOF and Tess OFF, High settings, AAA to MSAA4x, DX10, High Settings. Haven't tried getting into a big fight with the latter settings though.

What resolution? The game runs AA by default in DX11 (looks like a form of 2x, check my screenshots).

 

[Imp]

Holy crap, just played this for an hour, and it puts Crysis to shame on its ability to rape hardware. Using a 5770, i5 750 and 4 Gb Ram, and can only play smooth on high at 1680x1050, AAA and AFx4 (?).

I tried at 1920x1200 and high, 1920x1200 at very high, and it looks beautiful (motion blue overdone though) but runs like turd. Not happy about the lack of customization for effects without editing files. Ohwell, still looks great.

 

[Sylvanas]

Has anyone got a pic from taskmanager with an i7 with HT enabled? All 8 threads being used? I see a few pages back that it takes advantage of 4 cores on a Quad, this is probably the only game to do so properly.

 

[MrK6]

X-bit labs did a performance review of Metro 2033; it's pretty accurate comparing it with my experience. Enjoy: http://www.xbitlabs.com/articles/video/display/metro-2033.html

 

[Grooveriding]

X-bit labs did a performance review of Metro 2033; it's pretty accurate comparing it with my experience. Enjoy: http://www.xbitlabs.com/articles/video/display/metro-2033.html

It's worth noting in those benches you linked they have Dx11 options disabled. I run 5870 crossfire, cards clocked at 950 / 1300. There are definitely noticeable times where the game is not comfortable at 2560x1600.

I think this will improve when ATI releases new drivers with a profile for this game but it's a given this game will perform better on the 480.

Pretty obvious already this is NBA bog marketing game for fermi's release. You will be seeing benches of this game on nv hardware paraded everywhere.

 

[1h4x4s3x]

Wow, the game is scaling well.

 

[YBS1]

Has anyone got a pic from taskmanager with an i7 with HT enabled? All 8 threads being used? I see a few pages back that it takes advantage of 4 cores on a Quad, this is probably the only game to do so properly.

It doesn't look like it's taking advantage of hyper threading anyway. This wasn't in a particularly taxing area of the game, I don't know how hard it would work the cores in that case but it's obvious at least during the time period this shot was taken a Xeon clocked at ~3.8ghz is a bit overkill.

 

[RussianSensation]

X-bit labs did a performance review of Metro 2033; it's pretty accurate comparing it with my experience. Enjoy: http://www.xbitlabs.com/articles/video/display/metro-2033.html

They also said FSAA was 'useless' in the game since it blurred out the textures like so:

With AA
http://www.xbitlabs.com/misc/picture/?src=/images/video/metro-2033/Quality/1_vh4.jpg&1=1

Without AA - looks better
http://www.xbitlabs.com/misc/picture/?src=/images/video/metro-2033/Quality/1_vhigh.jpg&1=1

Anyone else agree with these findings?

 

[MrK6]

It's worth noting in those benches you linked they have Dx11 options disabled. I run 5870 crossfire, cards clocked at 950 / 1300. There are definitely noticeable times where the game is not comfortable at 2560x1600.

I think this will improve when ATI releases new drivers with a profile for this game but it's a given this game will perform better on the 480.

Pretty obvious already this is NBA bog marketing game for fermi's release. You will be seeing benches of this game on nv hardware paraded everywhere.

Of the DX11 options, tesselation is only worth about a 10-15% performance hit, not bad IMO. Advanced DOF kills FPS, so I usually leave it off. Either way, I'll agree there are some parts where FPS tanks into the 20s (generally while viewing some types of volumetric lights, check my screenshots), however performance was generally in the 50-60FPS range for me.

They also said FSAA was 'useless' in the game since it blurred out the textures like so:

With AA
http://www.xbitlabs.com/misc/picture/?src=/images/video/metro-2033/Quality/1_vh4.jpg&1=1

Without AA - looks better
http://www.xbitlabs.com/misc/picture/?src=/images/video/metro-2033/Quality/1_vhigh.jpg&1=1

Anyone else agree with these findings?

Yup, that's pretty much it. Seems like they use some kind of super-sampling AA as it does look worse. The thing is, you don't really need the 4x AA in the first place. Take a look at the screenshots I posted, which have only "AAA" antialiasing enabled. There aren't really any glaring jagged edges, and the IQ is fantastic. I'm actually using a screenshot from the game as my desktop now.

 

[toyota]

Of the DX11 options, tesselation is only worth about a 10-15% performance hit, not bad IMO. Advanced DOF kills FPS, so I usually leave it off. Either way, I'll agree there are some parts where FPS tanks into the 20s (generally while viewing some types of volumetric lights, check my screenshots), however performance was generally in the 50-60FPS range for me.
Yup, that's pretty much it. Seems like they use some kind of super-sampling AA as it does look worse. The thing is, you don't really need the 4x AA in the first place. Take a look at the screenshots I posted, which have only "AAA" antialiasing enabled. There aren't really any glaring jagged edges, and the IQ is fantastic. I'm actually using a screenshot from the game as my desktop now.

well the review on PCGH had tessellation and dof each giving basically the same performance hit.

 

[cbn]

Has any review site measured CPU core scaling yet?

Would a Gulftown with hyperthreading enabled give better FPS (on low graphical settings) than a Core i5 750 clocked at the same speed?

The developer claims this game will scale to 16 cores.

 

[RussianSensation]

Has any review site measured CPU core scaling yet?

Would a Gulftown with hyperthreading enabled give better FPS (on low graphical settings) than a Core i5 750 clocked at the same speed?

The developer claims this game will scale to 16 cores.

Xbitlabs review linked above shows little difference between Core i7 920 and 975. I doubt a GT would help.

This paints a far better picture:

CPU Speed Scaling
http://www.legionhardware.com/articles_pages/metro_2033_performance_guide,9.html

Core Scaling
http://www.legionhardware.com/articles_pages/metro_2033_performance_guide,6.html

Purely GPU limited game.

 

[toyota]

Has any review site measured CPU core scaling yet?

Would a Gulftown with hyperthreading enabled give better FPS (on low graphical settings) than a Core i5 750 clocked at the same speed?

The developer claims this game will scale to 16 cores.

it looks like they were exaggerating a bit. there seems to be no benefit using a quad over a dual at least when using moderate video card. http://www.legionhardware.com/articles_pages/metro_2033_performance_guide,6.html

now here is some cpu scaling with 5870 crossfire and Phenom X4 and i7 deliver the same performance no matter what clocks are used on the cpu. http://www.legionhardware.com/articles_pages/metro_2033_performance_guide,9.html

this game is certainly VERY gpu dependent so it appears any fast dual core will get the job done. I wish they would have compared duals to quads with the high end gpus though it likely wouldnt matter too much.